AI drug discovery is the corner of artificial intelligence with the highest stakes and the least room for hype: the products are molecules that go into human bodies, and the referee is a clinical trial that does not care about your demo. As a physician, here is how the field actually works, what has genuinely changed, and how to read the claims.
What AI actually does in discovery
Drug discovery has three bottlenecks where AI now earns its keep. Target identification: mining biological data to propose which protein to drug, the step where a mistake wastes a decade. Molecule generation: designing candidate compounds with the right binding, solubility and safety profile, a search across chemical space that generative models are legitimately good at. Triage: predicting which candidates will fail early, which matters because failure is the norm, roughly nine in ten drugs entering human trials never reach approval.
What has genuinely changed
The timeline. The industry norm from target to late-stage trials has historically run a decade or more; AI-native programs are now doing it in roughly half that. The field’s current bellwether is Insilico’s Rentosertib, the first drug with an AI-identified target and AI-generated molecule to enter Phase III, for idiopathic pulmonary fibrosis. Capital has noticed: health AI took 55 percent of health-tech funding last year, and big pharma now signs AI-discovery collaborations worth hundreds of millions in milestones.
How a physician reads the claims
Three filters separate signal from press release. First, what did the AI actually contribute: “AI-discovered” ranges from novel target plus novel molecule down to a screening step any pharma has run for years. Second, what phase is the evidence: preclinical success is a chemistry achievement; Phase III is a medical one; nothing counts until randomized human data exists. Third, speed is not efficacy: a faster pipeline that produces drugs which fail trials at the usual rate has changed economics, not medicine. The honest summary today: AI has provably compressed discovery timelines; it has not yet proven it makes better drugs. Rentosertib’s readout will be the first serious answer.
What to watch
Watch the Phase III readouts of AI-native drugs over the next two years, whether regulators begin distinguishing AI-designed candidates in guidance, and whether the evidence culture that scribes escaped gets enforced where it matters most: the molecule. A live example just cleared its first checkpoint: an AI-designed universal coronavirus vaccine passed Phase 1 safety trials.
Frequently asked questions
What does AI actually do in drug discovery?
AI assists at three bottlenecks: identifying which biological target to drug, generating candidate molecules with the right properties, and triaging which candidates are likely to fail early.
Has an AI-discovered drug ever reached Phase 3 trials?
Yes. Insilico Medicine’s Rentosertib, with an AI-identified target and AI-generated molecule, entered Phase III trials for idiopathic pulmonary fibrosis in 2026.
Does AI make drug discovery faster or just cheaper?
Primarily faster. AI-native programs are compressing the target-to-late-stage-trial timeline to roughly half the industry’s historical decade-plus norm.
Does a faster AI pipeline mean safer or more effective drugs?
Not necessarily. Speed compresses timelines; it does not change the failure rate in human trials, which is why Phase III data, not discovery speed, is the real test of the technology.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
